def __init__(self,
model_dim,
nheads,
ff_dim,
num_layers,
dropout=0.1,
mem_att_dropout=0.1,
mem_att=True,
mem_dim=None,
prop_dim=None):
super(TransformerStack, self).__init__()
layer = TransformerDecoderLayer(model_dim=model_dim,
nheads=nheads,
ff_dim=ff_dim,
dropout=dropout,
mem_att=mem_att,
mem_dim=mem_dim,
mem_att_dropout=mem_att_dropout)
self.layers = _get_clones(layer, num_layers)
self.num_layers = num_layers
self.norm = LayerNorm(model_dim)
self.tgt_mask = None
self.prop_dim = prop_dim if prop_dim is not None and prop_dim > 0 else None
if self.prop_dim is not None:
self.inp_proj = Sequential()
self.inp_proj.add_module(
"lin", Linear(model_dim + self.prop_dim, model_dim))
def __init__(self, d_model, N, heads, height, width, device):
super().__init__()
self.N = N
self.pe = LearnedPositionalEncoder(d_model, height, width, device)
self.layers = transformer._get_clones(
transformer.TransformerEncoderLayer(d_model,
heads,
dim_feedforward=d_model),
N,
)
def __init__(self,
decoder_layer,
decoder_final_layer,
num_layers,
norm=None):
super(TransformerDecoder, self).__init__()
self.layers = _get_clones(decoder_layer, num_layers - 1)
self.final_layer = decoder_final_layer
self.num_layers = num_layers
self.norm = norm
def __init__(self, decoder_layer, num_layers, norm=None):
super(TransformerDecoderCustom, self).__init__()
self.layers = _get_clones(decoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm
def __init__(self, encoder_layer, num_layers, norm=None):
super(TransformerEncoder, self).__init__()
self.layers = _get_clones(encoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm
def __init__(self, decoder_layer, num_layers, norm=None):
super().__init__()
self.layers = _get_clones(decoder_layer, num_layers)
self.num_layers = num_layers
self.norm = norm